Monetary Policy and the Term Structure of Interest Rates in Japan

By

R. Anton Braun The University of Tokyo

and

Etsuro Shioji Yokohama National University

April 26, 2004

We would like to thank the editor and two anonymous referees for their valuable comments. We also thank Bob Conroy and Nobuya Takezawa for assistance in collecting the data and seminar participants at the Bank of Japan, the CIRJE/CEPR/EIJS/NBER Japan Project Meeting, Duke University, Keio University, North Carolina State University, RIEB at Kobe University and Yokohama National University for their comments. The first author received support for this project from the Foundation for International Education and the Ministry of Education of Japan. The second author would like to thank financial assistance from the Nikkei Foundation, the Seimei Foundation, and the Ministry of Education of Japan.

2

Abstract This paper uses Japanese data to investigate the relationship between monetary policy and the yield curve. We find that the response of the yield curve depends in an important way on the maintained hypothesis about how monetary policy affects the economy. Under the liquidity effect maintained hypothesis monetary policy only has transient effects on the yield curve. Under the costly price adjustment maintained hypothesis, however, monetary policy has large and persistent effects on yields of all maturities.

3

1 Introduction

This paper investigates the relationship between the Japanese yield curve and monetary

policy. In the 1980’s and 1990’s average bond yields have risen from 5% to 8% and then

fallen to 2% and the slope of the yield curve has swung from positive to negative to

positive. We are interested in understanding the contribution of monetary policy to these

movements in the yield curve.

One motivation for our interest is Japan’s recent experience. In spite of massive increases

in monetary base and a zero nominal interest rate, economic growth has remained low and

deflationary pressure has not abated. These events are raising new questions about the

effectiveness of monetary policy under a zero nominal interest rate policy. (Eggertson

and Woodford 2003) argue that a monetary authority can still influence economic activity

when nominal interest rates are zero by taking actions that affect market expectations

about the future time path of variables such as interest rates, inflation or exchange rates.

One way to assess the ability of a central bank to affect expectations is to look

retrospectively and ascertain the extent to which previous monetary policy surprises have

affected bond yields of different maturities. If monetary policy is indeed a potent tool for

altering expectations then this should show up in the responses and variance

decompositions of medium and long-term bonds yields to suitably identified shocks to

monetary policy.

4

In order to isolate the effects of monetary policy on the yield curve we must first identify

monetary policy shocks. Our strategy for identifying monetary policy combines zero

restrictions as in (Christiano, Eichenbaum, and Evans 1996), (Bernanke and Mihov 1996),

(Leeper, Sims and Zha 1996), (Miyao 2002) and (Shioji 1997) with sign restrictions on

the impulse response functions as in (Faust 1999) and (Uhlig 1999). An advantage of our

empirical strategy is that it is straightforward to investigate the robustness of any

conclusions to the maintained assumptions about how monetary policy affects the

macro-economy.

We consider two distinct maintained hypotheses. The liquidity effect hypothesis

maintains that a surprise tightening in monetary policy increases short-term nominal

interest rates, and lowers output, prices, and monetary aggregates. This hypothesis

reflects the consensus view about how monetary policy affects the U.S. economy [See e.g.

the recent survey article by (Christiano, Eichenbaum and Evans 1999)]. We also consider

the costly price adjustment hypothesis. This hypothesis maintains that a surprise

tightening in monetary policy lowers interest rates, lowers money supply, lowers output

and lowers prices. It is consistent with the implications of costly price adjustment models

with monopolistic competition as in: (Rotemburg 1996), (Christiano, Eichenbaum and

Evans 1997), (Ireland 1997), and (Aiyagari and Braun 1998). (Braun and Shioji 2002)

find that Japanese data are more consistent with the costly price adjustment hypothesis.

Here we report results under each of the two maintained hypotheses in order to compare

5

their implications for the Japanese yield curve.

The choice of maintained hypothesis has important implications for the interaction of

monetary policy and the yield curve. Under the liquidity effect hypothesis innovations in

monetary policy have highly transient effects on short-term interest rates and the slope

and curvature of the yield curve. Moreover, monetary policy shocks only account for a

small fraction of the long-run variance in yields. Under the costly price adjustment

hypothesis, in contrast, there is a rich set of interactions between monetary policy and the

yield curve. Monetary policy shifts the level of the yield curve and produces large

hump-shaped responses in yields of all maturities. Monetary policy also accounts for a

substantial fraction of the long-term variance in long-term yields.

Our analysis is related to recent work by (Ang and Piazzesi 2003) and (Evans and

Marshall 2001). (Ang and Piazzesi 2003) consider the role of alternative macroeconomic

shocks in explaining movements in U.S. Treasury yields using an affine model of the

term structure and find that economic activity accounts for only a small fraction of the

variance in long term bonds. (Evans and Marshall 2001), in contrast, use a common

factor model of the term structure and identify a variety of macroeconomic shocks. They

find that demand shocks account for a significant fraction of the variance in long term

bonds. Our work complements these papers in several ways. We describe how the

implications of monetary policy for the yield curve vary across alternative maintained

hypotheses about the economic effects of monetary policy shocks. We also consider the

6

role of financial shocks in explaining movements in macroeconomic variables. Finally

we investigate these issues for Japan, which has a different institutional and economic

environment from the United States.

2 The Model

2.1 Reduced form Vector Auto-Regression

The reduced form econometric model consists of a vector auto-regression or VAR

0 1 1 2 2t t t J t J tx C C x C x C x u− − −= + ⋅ + ⋅ + + ⋅ + (1)

where J denotes the number of lags, tx is an (mx1) vector of variables and tu is an (mx1)

vector of disturbances. Denote the covariance matrix of tu as Σ . The baseline VAR

specification includes 6 lags of monthly data on four macroeconomic variables: the

Consumer Price Index less food (CPI), Industrial Production (Y), the monetary base

adjusted for reserve requirements (M), and the one month TIBOR rate (R). The sample

period is October 1987 through May 1999. All variables are expressed in log-levels with

the exception of R, which is expressed in levels. 1 The price level and industrial

production are included because they summarize the two principal objectives of monetary

policy: promoting price stability and stabilizing fluctuations in output. The Monetary

base and the one month TIBOR rate are both included to help discriminate between the

implications of the liquidity effect hypothesis and the costly price adjustment hypothesis.

7

Under the liquidity effect hypothesis narrow money and the short rate move in opposite

directions in response to an innovation in monetary policy. The costly price adjustment

hypothesis, in contrast, implies that these two variables move in the same direction in

response to an innovation in monetary policy. This identification issue is discussed in

more detail in Section 2.

A number of other variables have also been considered elsewhere in the literature. Most

prominently commodity prices have been shown to render U.S. data more consistent with

the liquidity effect hypothesis [see e.g. (Sims 1992) and (Christiano, Eichenbaum and

Evans 1996)]. Given the important role that exports play in the Japanese economy, the

exchange rate may also be either an important information variable for the Bank of Japan

or possibly a target of monetary policy. Finally, Japan imports most of its oil and

economic activity may be sensitive to fluctuations in the price of oil. To explore these

possibilities we also report results below in which the baseline list of macroeconomic

variables is augmented to include alternatively the commodity price index (PCOM), the

yen/$ exchange rate (YENDOL) or an oil price index (POIL).

To complete the list of variables two common factors (F1, F2) are included that in

conjunction with R summarize the dynamics of the yield curve. Previous work has found

that the yield curve is well-summarized by three factors that respectively shift its level, its

slope and its curvature [see e.g. (Litterman and Scheinkman 1988), (Singleton 1994) and

(Hiraki, Shiraishi and Takezawa 1996)]. We also assume a three-factor model of the

8

Japanese yield curve. However, in contrast to the previous literature, the first factor is

taken to be the one-month TIBOR rate. The remaining two common factors, F1 and F2

are estimated by principal components. This insures that they are by construction

orthogonal to the one-month rate and to each other.2

2.2 identification of monetary policy

We assume that the disturbances are driven by m structural shocks that are mutually

orthogonal:

1t tu P ε−= ⋅ , (2)

where P is a (mxm) matrix, and the covariance matrix of tε is a (6x6) identity matrix.

Suppose further that the monetary policy shock is the fourth element in the vector tε .

Under these assumptions, identifying monetary policy amounts to determining the values

of the elements in the fourth column of the matrix P-1.

Our strategy for identifying monetary policy- that is the elements of this fourth column- is

Bayesian. Following (Uhlig 2001) we assume a diffuse normal/Wishart prior over

0 1([ , ,.., ], )JC C C Σ that is multiplied by an indicator function. This indicator function

allows us to indirectly impose prior restrictions that are difficult to impose directly on the

parameters of the model. The value of this indicator function is determined by the

intersection of two events. First, throughout the whole analysis attention is limited to a

block recursive identification structure. Second, the value of the indicator function is

9

determined according to whether or not a set of sign restrictions on the impulse response

to innovations in monetary policy are satisfied. These sign restrictions embody a

particular prior about how monetary policy affects the economy. We consider each of

these events in turn.

Recursive restrictions

We start by partitioning the model variables into three groups. For the baseline economy

prices and output are assigned to the first partition, monetary base and the nominal

interest rate are assigned to the second partition and the two yield curve common factors

are assigned to the third partition. Assume further that 1P− is block triangular:

111

1 1 121 22

1 1 131 32 33

0 00

PP P P

P P P

−

− − −

− − −

⎛ ⎞⎜ ⎟

= ⎜ ⎟⎜ ⎟⎝ ⎠

(3)

Under these assumptions shocks to variables in the first partition affect all variables

contemporaneously, shocks to variables in partition 2 affect only variables in the second

and third partition, and shocks to variables in the third partition have no contemporaneous

impact on variables in either the first or second partitions.

This block recursive structure partitions the time t information set of the monetary

authority into two parts: variables that the monetary authority observes prior to setting

current period monetary policy and variables that the monetary authority doesn’t observe

10

contemporaneously. In particular, we assume that the monetary authority observes the

current shocks to output and prices prior to setting monetary policy but, does not observe

the period t shocks to either financial sector variable. The former restriction is relatively

common in the literature [See e.g. (Christiano and Eichenbaum 1992), (Bernanke and

Blinder 1992), and (Gertler and Gilchrist 1994)] and is also consistent with the

implications of dynamic general equilibrium models of money such as (Christiano,

Eichenbaum and Evans 2001).

The latter restriction is imposed to mitigate the risk of an identification problem raised by

(Leeper, Sims and Zha 1996). They give an example of a monetary policy feedback rule

that reacts to multiple current period interest rates and show that this rule can induce

indeterminacy of equilibrium and can also render identification infeasible.

Our assumptions also imply that demand for the monetary base does not respond

contemporaneously to shocks in the two common factors. We think of demand for base

money as coming from three main sources; exchange credit that facilitates trade among

firms as in (Kahn and Roberds 2002); transactions demand by households- carrying cash

can be convenient if ATM machines are not nearby or closed3; and money is needed to

settle tax payments with the government. We are assuming that each of these three

demands for money is insensitive to current shocks to the yield curve.

Given this block recursive structure, identification of the innovation to monetary policy

11

involves pinning down the coefficients in 122−P and 1

32P− . Block recursive identification

schemes have several convenient properties. One of them is that identification of 122−P

does not depend on the values of P-1 in partitions 1 and 3. In particular, 122−P can be

derived without referring to any of the elements of the other partitions. A second property

is that

1 132 22P P− −= Θ⋅ (4)

where Θ is a two by two matrix that is uniquely determined from Σ so that once 122−P is

identified, 132P− is also identified.. These properties can be ascertained directly using the

fact that 1 1 'P P− −Σ = and by imposing the block recursive zero restrictions. It follows that

identification of sector 2 shocks can proceed without making any further assumptions

about how the shocks in the other two sectors are identified.

Sign restrictions

We turn next to describe how we define events that satisfy the block recursive restrictions

and thereby identify 122P− . The general strategy is to use simulation methods to produce a

pseudo-random sequence of VAR parameters, deduce a sequence of 122P− ’s by imposing

the block recursion structure and then to use these objects to construct a pseudo-random

sequence of impulse response functions. Given this sequence of impulse response

functions rejection methods are used to impose the sign restrictions implied by a

particular prior about how monetary policy affects the economy.

12

We start by taking k1 random draws from a Normal/Wishart family whose parameters are

given by the estimated reduced form VAR coefficients, 0 1ˆ ˆ ˆ ˆ[ , ,... ]JC C C C≡ , and the

estimated covariance matrix of disturbances, Σ . Given a draw from the distribution of

the VAR parameters, the recursive identification restrictions imply a particular 122−P . Our

interest centers on the second column of 122−P because it corresponds to a shock to

monetary policy. The elements of 122−P are related to Σ , in the following way:

1 1 122 21 11 21 22 22' 'P P− − −Ω ≡ Σ −Σ ⋅Σ ⋅Σ = . (5)

Denote the eigen-values of Ω as 1µ and 2µ , and the corresponding normalized

eigenvectors by 1v and 2v . Then, a, the second column of 22

1P− has the following

representation:

2

1i i i

i

a vα µ=

= ⋅ ⋅∑ (6)

where,

2 21 1 1α α+ = .

Uhlig (2001) shows that the 'sα defined in this way provide a complete characterization

of the set of a’s that satisfy 1 122 22 'P P− −Ω = .

The representation given by (6) also implies that an innovation to monetary policy is only

identified up to a one-dimensional continuum that is indexed by 1α . A particular prior

13

about how monetary policy affects the economy defines a subset of this one-dimensional

continuum. To find this subset we draw k2 random 1α ’s from a uniform [-1,1] distribution,

setting 2α so that the squared α ’s sum to one. At this point we have a completely

identified system and we can compute the impulse response function to monetary policy.

Rejecting draws (events) that violate the sign restrictions imposes a particular prior about

how monetary policy affects the economy.

We consider two distinct sets of sign restrictions. Each set of sign restrictions corresponds

to a competing prior about how monetary policy affects economic activity. The first prior

is referred to as the Liquidity Effect Hypothesis. Under this hypothesis we assume that

(1) the response of the price level is negative in a majority of the first 7 months following the arrival of a contractionary shock to monetary policy; 4

(2) the response of output is negative in a majority of the first seven months following the arrival of the shock;

(3) the response of the monetary base is negative in a majority of the first six months;

(4) the response of the one-month rate is positive in a majority of the first six months.

The liquidity effect maintained hypothesis is designed to reflect the consensus view about

how monetary policy affects the economy. (Friedman 1968) suggests that liquidity

effects might last for up to a year. And results reported in the survey article by (Christiano,

Eichenbaum and Evans 1999) are consistent with these restrictions with the possible

exception of the price level.5

The second maintained hypothesis is referred to as the Costly Price Adjustment

14

Hypothesis. The costly price adjustment hypothesis consists of the following sign

restrictions:

(1) the response of the price level is negative in a majority of the first 7 months following the arrival of an innovation to monetary policy;

(2) the response of output is negative in a majority of the first seven months following the arrival of a shock to monetary policy;

(3) the response of the monetary base is negative in a majority of the first six months following the arrival of a shock to monetary policy;

(4) the response of R is negative in a majority of the first six months following the arrival of a shock to monetary policy.

This second hypothesis is consistent with the implications of monopolistically

competitive costly price adjustment models such as (Rotemberg 1996), (Christiano,

Eichenbaum and Evans 1997), (Ireland 1997) and (Aiyagari and Braun 1998). In these

models a surprise contraction in monetary policy reduces output in the short-run because

prices are now high relative to future periods, gradually lowers prices, lowers the growth

rate of monetary aggregates and lowers nominal interest rates due to an expectation that

inflation in future periods will fall. While these responses are consistent with some

leading sticky price models of money, not all costly price adjustment models have this

property. (Christiano, Eichenbaum and Evans 2001), for example, develop a model with

costly price adjustment in which the responses of the economy to shocks in monetary

policy are consistent with the liquidity effect hypothesis. Instead, it is probably best to

view this hypothesis as reflecting effects that are plausible in the sense that they are the

dominant effect in the dynamic general equilibrium models listed above.

15

Our algorithm places restrictions on both the choice of P and the parameters of the

reduced form VAR. This approach to identification differs from the standard approach in

this literature, which seeks to identify P conditional on a particular choice of the model

parameters. In Section 3 we compare and contrast the two approaches.

Observe also that the form of the sign restrictions is based entirely on the sign and assigns

no weight to the magnitude of the responses. On the one hand, count restrictions better

reflect the nature of the consensus about how monetary policy affects the economy.

Statements of the consensus perspective focus more on signs than magnitudes [see e.g.

(Christiano, Eichenbaum and Evans 1999)]. On the other hand, it is possible that sign

counts rule out identifications that, for instance, produce big but highly transient liquidity

effects. Below we will show that the empirical results are robust to the choice of imposing

the hypotheses as sign count restrictions or, alternatively, restrictions on the mean

responses over the first 6-7 periods.

3 Estimation results

3.1 Estimation and simulation of the model

Estimation and identification of the model proceeds in three steps. First, the common

factors of the term structure are estimated. Second, the VAR is estimated and third,

monetary policy is identified using the simulation based rejection method described

above.

16

The underlying data consists of zero-coupon equivalent yields with 11 different

maturities. Common factors are constructed from the yield data in the following manner.

Each of the eleven yields was regressed on the 1-month rate. Then a principal component

analysis was performed on the eleven residual series. This analysis indicated that the

components corresponding to the two largest eigen-values explain 95.1% of the total

variation in the eleven residual series. As a check of the ability of the 1-month rate and the

two common factors to summarize movements in yields of alternative maturities, each of

the 11 yields was regressed on these three variables. The R square from the regressions

was greater than 0.990 in all cases.

The weights the 1-month rate and the two common factors receive on each yield are

reported in Table 1. Inspection of Table 1 indicates that the patterns of the estimated

coefficients do not bear much resemblance to the level, slope and curvature factors. The

magnitude of the first factor (1-month TIBOR rate) falls by nearly half as the maturity of

the yield rises from 2 months to 10 years. In addition, the sign changes that one would

associate with slope or curvature factors are not present in the coefficients for the first

factor. The coefficients for the second factor show a change of sign but the magnitude of

the 2 month yield and 120 month yield are not very close in absolute value.

Given this difficulty in interpreting these factors it is perhaps helpful to explain why we

chose to proceed in this manner. The main consideration is parsimony. Since the VAR

already includes the one-month rate, treating the one-month rate as a common factor and

17

adding two additional factors summarizes the same information as a conventional three

factor model and reduces the number of estimated coefficients by 92.

When estimating the VAR we treat the two estimated common-factors as data as in e.g.

(Bernanke and Boivin 2003). Moreover, responses of yields also condition on the

weights reported in Table 1. Our two step procedure understates the inherent uncertainty

in the model. An alternative strategy would be to directly include additional yields in the

VAR instead of the two common factors. We chose not to pursue this alternative strategy

because of a concern about the numerical stability of the estimates. Yields of different

maturities are highly correlated. For our dataset, the contemporaneous correlation

coefficients range between 0.95 and 0.99. In such a situation including multiple yields in

the same VAR can have a large effect on the condition number of the data matrix that one

inverts in the course of OLS estimation of the model. 6 For Japanese data this problem is

compounded by the fact that the sample period over which yields are available is

relatively short and comovements in the data are strongly influenced by the rise and fall of

stock and land prices in the late 1980’s and 1990’s.

With the common factors in hand, the VAR is estimated and the restrictions of the

liquidity effect maintained hypothesis are imposed. Following the discussion in Section 2,

the first step is to take a pseudo-random draw from the posterior distribution of the VAR

coefficients. Next we take 100 draws from the free elements of 122P− and check to see if

they satisfy the sign restrictions for the costly price adjustment hypothesis. For successful

18

draws we tabulate partial sums of the impulse response functions and variance

decompositions. This process is repeated for 500 draws from the posterior distribution of

the VAR coefficients producing a total of 50,000 trials. Finally, the averages and standard

deviations of the impulse responses, and variance decompositions are calculated. This

same set of procedures is then repeated for the costly price adjustment maintained

hypothesis.

3.2 What are the effects of monetary policy on economic activity and yields under

the two hypotheses?

Impulse responses: liquidity effect hypothesis

The left panels of Figure 1 and 2 report impulse responses to a tightening in monetary

policy under the liquidity effect maintained hypothesis. Figure 1 contains responses for

the VAR variables and also LEVEL, SLOPE, CURVATURE and the real interest rate.

Figure 2 shows responses of 2-month, 12-month and 60-month yields and term premia.

The impulse responses reported in these and all other figures are the average of the

impulse responses across valid draws. The other two lines in each figure are respectively

an upper two standard deviation error band and a lower two standard deviation band

across valid draws. The responses for prices, output, and monetary base are expressed in

percent (.01=1%) and the other variables are expressed as annualized basis points.

The results reported in Figure 1 are based on an overall total of 486 successful draws or

19

about 1% of the 50,000 trials. Only 8.4% of the outer-loop draws from the posterior

produce one or more valid draw. And on average each successful draw from the posterior

distribution produces about 12 valid 1α ’s.

Under the liquidity effect maintained hypothesis monetary policy has large but highly

transient effects on economic activity. The response of output is large and immediate. A

one standard deviation (10 basis point) tightening in the 1-month rate produces a peak

decline in the level of output of 20/100th’s of a percent in the second month following the

tightening. However, neither the output response nor the 1-month rate response is very

persistent. Both variables are within 2 standard deviations of zero within four months

after the shock arrives and damp quickly. The two common factors also respond

significantly on impact but then damp quickly thereafter. Prices don’t respond much in

early periods but gradually fall over time. None of the price responses are more than two

standard deviations from zero. The monetary base is the only variable with a persistent

response that is precisely estimated. Monetary base drops on impact and is more than two

standard deviations from zero in most periods.

Figure 1 also reports the responses of LEVEL, SLOPE, CURVATURE and the real

1-month interest rate. The definitions of LEVEL, SLOPE, and CURVATURE follow the

example of (Ang and Piazessi 2003). LEVEL is the average response of the 1-month rate,

the 12-month yield and the 60-month yield. SLOPE is the response of the 60-month yield

minus the 1-month rate and CURVATURE is defined as the response of the 1-month rate

20

plus the 60-month rate minus the twice the 12-month rate. Finally, the real interest rate is

the 1-month rate net of the expected inflation rate. A comparison of LEVEL, SLOPE and

CURVATURE with respectively the 1-month rate, F1 and F2 indicates that the shapes and

statistical significance of the responses are very similar. The biggest difference concerns

the impact response of LEVEL and the 1-month rate. Monetary policy innovations induce

a larger impact response in the 1-month rate. This difference may explain why the

coefficients on the 1-month rate in Table 1 are declining in maturity.

The sign of the response of the real interest rate is large but imprecisely estimated. On

impact, the response is 18 basis points and rises to a maximum of 23 basis points. In

subsequent periods it continues to fluctuate but with no distinct pattern. The error bands

are also very large though indicating that the precision of these estimates is very low.

Figure 2 reports the response of the yield curve, term premia and two standard deviation

confidence bands to the same innovation for maturities of 6-months, 12-months and

5-years.7 Term premia are calculated as departures from the Expectations Hypothesis as

in (Evans and Marshall 1998).8

Figure 2 has three noteworthy features. First, the magnitude and statistical significance

of the impact responses fall with maturity. While the response of the 6-month yield is

about 6 basis points and about two standard deviations away from zero, the response of

the 5-year yield is less than 4 basis points and within one standard deviation of zero.

Second, the effect of monetary policy shocks on the yield curve under the Liquidity Effect

21

hypothesis disappears after about 7 months. Third, term premia responses are small,

transient and similar to the responses of yields. This follows from transient nature of the

1-month rate responses. Averages of future 1-month rate responses are about zero. These

results imply that innovations in monetary policy only have transient effects on short end

of the yield curve under the liquidity effect maintained hypothesis.

Impulse responses: costly price adjustment hypothesis

Figure 1 also reports impulse response functions for the costly price adjustment

hypothesis. The results are based on the same number of replications as for the liquidity

effect hypothesis. Here the total number of successful draws rises to 8532 or about 17%

of the total draws. Over 66% of the draws from the posterior distribution of the

parameters produce at least one valid identification and the average number of valid 1α ’s

per successful draw from the posterior distribution of parameters exceeds 25.

These impulse responses correspond to a contractionary surprise in monetary policy, that

is, a monetary policy shock that lowers output. By comparing the left and right panels of

Figure 1 we see that the costly price adjustment maintained hypothesis produces

responses that are larger in magnitude and more persistent. The response of output

(industrial production) is bowl-shaped and falls by a maximum of 0.04% in the twelfth

month following a one standard deviation (12 basis point) decline in the one-month rate.

Prices and the one-month rate also fall persistently. These responses also exhibit higher

precision than under the alternative liquidity effect maintained hypothesis. Price

22

responses are about two standard deviations below zero from month seven and on. And

the responses of output are more than two standard deviations below zero in 6 out of the

first 12 months following the shock.

Once again the responses and precision of LEVEL, SLOPE and CURVATURE are quite

similar to the responses of respectively the one-month rate, F1 and F2. Now monetary

policy shocks have potent dynamic effects on the level of the yield curve. SLOPE first

falls and then rises and curvature goes up in early periods.

The response of the real interest rate is also now somewhat smaller but more persistent. It

declines by 21 basis points on impact and gradually rises to -6 basis points by month 24.

As before, the real interest rate responses are imprecisely estimated.

Responses of yields and term premia to an innovation in monetary policy are reported in

the right panel of Figure 2. From the perspective of the costly price adjustment hypothesis

monetary policy has big and persistent effects on yields of all maturities. All yields have

bowl-shaped responses that bottom out at about month seven at 18-20 basis points. The

responses are also more precisely estimated as compared to the left panel of Figure 2.

Yields with maturities of 6 and 12-months are both more than two standard deviations

below zero for more than a year and the 5-year yield is more than two standard deviations

below zero for 9 months. The increased persistence in the 1-month rate response also

produces persistent responses in term premia. Twenty-four months after the shock term

premia for all three yields are still between 5 to 6 basis points below zero.

23

Variance decompositions

Table 2 reports the fraction of variance in output, prices, monetary base, interest rates, the

two common factors and the 12-month and 5-year yields accounted for by innovations in

monetary policy. The numbers in parentheses are standard errors. Under the liquidity

effect maintained hypothesis monetary policy explains about 29% of the variance in the

one month rate and 58% of the variation in monetary base at step 1. Monetary policy

shocks, however, are not important sources of variation in output, prices or the yield

curve. Only a maximum of about 5% of the variance in prices, 4% of the variation in

output and 6-7% of the variance in yields is attributed to monetary policy at any forecast

horizon.

Under the costly price adjustment maintained hypothesis monetary policy continues to be

an important source of variation in monetary base and short-term rates. However,

monetary policy is also more important for understanding movements in other variables.

Now, the fraction of variance in output and prices accounted for by innovations in

monetary policy is bigger at longer forecast horizons. At step 60, for instance, monetary

policy explains 16% of the variance in prices and 16% of the variance in output. Under

the costly price adjustment maintained hypothesis monetary policy is also important for

understanding movements in yields of all maturities at all forecast horizons. At step 12,

monetary policy accounts for 40% of the variation in the 12-month yield and 35% of the

variation in the 5 year yield. And at step 60 monetary policy explains 28% of the variation

24

in both yields.

3.2 Is Japanese data more consistent with the liquidity effect hypothesis or the costly

price adjustment hypothesis?

A possible concern about our method for identifying monetary policy is that good draws

are sufficiently rare that they may be coming from the tail of the posterior distribution of

parameter coefficients and in this sense may not be representative of the data. Some of

the diagnostics presented above suggest that this is more of an issue for the liquidity effect

hypothesis. For instance, only 8% of the draws from the posterior distribution of the VAR

parameters produce one or more draws that are consistent with this hypothesis. For the

costly price adjustment hypothesis the corresponding figure is 66%.

In order to investigate this issue further we conditioned on the estimated parameters of the

VAR as is the convention in the structural VAR literature and took 50,000 draws from 1α .

For the costly price adjustment hypothesis the fraction of successful draws was 28%.

Moreover, the impact responses of monetary base and the 1-month rate are precisely

estimated.9 However, for the liquidity effect hypothesis no successful draws were found.

We then increased the number of replications to 100,000 and still found no successful

draws.10 We draw three conclusions from these results. First, the costly price adjustment

maintained hypothesis is more consistent with Japanese data than the liquidity effect

alternative.11 Second, if one conditions on the estimated parameters of the VAR when

25

conducting a specification search for monetary policy, it is likely to be quite difficult to

find any specification that is consistent with the liquidity effect hypothesis. Third, for

those who place strong faith in the liquidity effect hypothesis, allowing for parameter

uncertainty in the VAR coefficients provides a way to reconcile this prior with Japanese

data. Based on these results and in order to conserve space, the remainder of the paper

will just report results for the costly price adjustment maintained hypothesis.

4 Additional implications of the costly price adjustment maintained

hypothesis

4.1 Are macroeconomic shocks important sources of variation in the yield curve?

(Ang and Piazessi 2003) using U.S. Data find that macroeconomic shocks are important

for explaining movements in short-term yields but not important for understanding

movements in long-term yields. (Evans and Marshall 2001), in contrast, find that

macroeconomic shocks explain most of the long-run movements of yields. Variance

decomposition results reported in Table 3 allow us to assess the role of macroeconomic

shocks in explaining the dynamics of Japanese yields. This table reports the variance

decomposition of the 12-month and 60-month yield using the baseline costly price

adjustment specification. At short-forecast horizons, monetary policy and the two

common factors explain most of the variance in the yield curve. At step 1, about 56% of

the variance in the 12-month yield and 73% of the variance in the 5-year yield is

explained by the combination of the two common factors. At longer horizons, though the

26

macroeconomic shocks are more important. At step 60, the combination of output and

prices accounts for about 38% of the variance in the 12-month yield and about 25% of the

variance in the 5-year yield. Monetary policy explains another 30% of the movements in

these two yields. Overall, our results for Japan are more consistent with the findings of

(Evans and Marshall 2001). However, monetary policy is more important in Japan for

understanding long-run movements in long-term yields. This finding is consistent with

some other facts about the Bank of Japan. As compared to the Federal Reserve, the Bank

of Japan holds a much larger fraction of long-term bonds on its balance sheets. About

60% of Japanese Monetary Base is backed by long-term government bonds. In addition,

the overall size of the Bank of Japan’s balance sheets is substantially larger than those of

the Federal Reserve.

4.2 Are financial shocks important sources of variation in macroeconomic activity?

(Estrella and Hardouvelis 1991) find that the slope of the yield curve is a predictor of

future economic activity in the 1970’s and 1980’s in U.S. data. Table 4 reports

decompositions of the variance of output and prices under the costly price adjustment

maintained hypothesis. Financial shocks don’t explain much of the variance in either

prices or output at short and medium (1 year) horizons. At long horizons financial sector

shocks are somewhat more important. F1 and F2 explain a combined fraction of 17% of

the variance in prices and 26% of the variance in output at step 60.

4.2 Are the results robust?

27

Here we explore the robustness of the results in our baseline specification along three

dimensions: the choice of variables, the specification of the nominal interest rate and the

method of imposing the sign restrictions.

Figures 3 and 4 report the impulse responses for specifications in which we add a seventh

variable to the VAR. Our interest here is determining whether our conclusions depend in

an important way on the choice of variables. The first panel of each figure reports results

for the baseline VAR variables plus PCOM, a commodity price index ordered third in the

first block. The center panel reports results for the baseline variables plus POIL, the price

of oil ordered third in the first block and the right panel reports results for the baseline

variables with the yen/$ exchange rate (YENDOL) ordered first in the third block of

variables. A comparison of these results with the baseline specification results reported in

Figures 1 and 2 suggests that adding these variables does not have much of an effect on

the results reported above. The main difference is that including the yen/$ exchange rate

dampens the response of the real interest rate. The implications for the yield curve though

are virtually the same as the baseline specification.

Given that nominal interest rates in Japan are very low and simple arbitrage arguments

suggest that zero is a lower bound, it is also interesting to consider ways to impose this

restriction on the specification and investigate whether such restrictions affect our results.

We considered two ways to impose this restriction. First, we add an additional test to the

Monte Carlo simulations that requires that the unconditional mean of the 1-month rate

28

was restricted to be non-negative. This doesn’t rule out negative realizations of the

nominal interest rate but does rule out the possibility of a negative average nominal

interest rate. When we do this the number of successful draws falls to 3450. However,

imposing this restriction has virtually no impact on the impulse response functions.12

Second, we considered the following nonlinear transformation when the simulated mean

of the nominal interest rate was less than 1% per annum:

, 1%

ln( ) 1 , 1%t t

tt t

R RR

R R>⎧

= ⎨ + <=⎩ (7)

where the one month rate is expressed as an annualized percentage. This transformation

rules out negative realizations of the 1-month rate in simulations where the mean of the

1-month rate is less than 1%. The fraction of successful draws based on this specification

is 19% and the impulse response functions are also close to those from the baseline

specification.13

Finally, we performed runs in which the mean response during the first 6 periods was

restricted instead of the sign. In this case, the fraction of successful draws for the baseline

specification rises to 24% but the impulse responses are again qualitatively and

quantitatively similar to the baseline case.14

5 Concluding remarks

In this paper we have considered the effect of monetary policy on the Japanese yield

curve. We have found that the effectiveness of monetary policy in affecting future

29

expectations depends importantly on the maintained hypothesis about how monetary

policy affects the economy. According to the liquidity effect hypothesis monetary policy

is an ineffective tool at manipulating expectations about future nominal interest rates.

Neither long-term bonds nor term premia respond persistently to monetary policy shocks.

An entirely different picture emerges under the costly price adjustment maintained

hypothesis. According to this perspective shocks to monetary policy in the 1980’s and

1990’s were important sources of variation in bonds of all maturities. Moreover, this

hypothesis implies that the monetary authority has lots of ammunition. Monetary policy

surprises that act to drive up short rates will alter expectations about future interest rates

by shifting the level of the yield curve up in a persistent way and thereby stimulate the

Japanese economy for a period of about two years.

In future work we plan to consider a broader array of assets and undertake a more

complete identification of other macro shocks. This will allow us to quantitatively assess

the role of various macroeconomic shocks in explaining events such as the collapse of the

Japanese asset price boom in 1990.

Data Appendix

Our data for the Consumer Price Index and Industrial Production (both seasonally

adjusted, 1990 average = 100) come from the Nikkei NEEDS Macroeconomic Database.

Our data for the Monetary Base (adjusted for reserve requirement ratio changes, monthly

average, seasonally adjusted) were downloaded from the Bank of Japan’s web site.

30

The yield data used to estimate the two common factors consists of TIBOR rates with

maturities of 2, 3, 6, 9 and 12 months and off-shore swap rates with maturities of 24, 36,

48, 60, 95, and 120 months. All yields are expressed on a zero coupon equivalent basis.

The source of this data is Datastream.15

References

Aiyagari, S. Rao, and R. Anton Braun. (1998). “Some Models to Guide Monetary

Policymakers.” Carnegie-Rochester Conference Series on Public Policy 48, 1-42. Ang, Andrew and Monika Piazzesi. (2003). “A No-Arbitrage Vector Autoregression of

Term Structure Dynamics with Macroeconomic and Latent Variables.” Journal of Monetary Economics 50, 745-787.

Bernanke, Ben S., and Alan S. Blinder. (1992). “The Federal Funds Rate and the

Channels of Monetary Transmission.” American Economic Review 82, 901-21. Bernanke, Ben S., and Jean Boivin. (2003). “Monetary Policy in a Data-Rich

Environment.” Journal of Monetary Economics 3, 497-720. Bernanke, Ben S., and Ilian Mihov. (1998). “Measuring Monetary Policy.” Quarterly

Journal of Economics 108, 869-902. Braun, R. Anton and Etsuro Shioji. (2002). “Monetary Policy and Economic Activity in

Japan and the United States.” Unpublished manuscript. Christiano, Lawrence, and Martin Eichenbaum. (1992). “The Liquidity Effects and the

Monetary Transmission Mechanism.” American Economic Review 82, 346-53. Christiano, Lawrence, Martin Eichenbaum, and Charles Evans. (1996). “The Effects of

Monetary Policy Shocks: Some Evidence from the Flow of Funds.” Review of Economics and Statistics 78, 16-34.

Christiano, Lawrence, Martin Eichenbaum, and Charles Evans. (1997). “Sticky Price and

Limited Participation Models of Money: A Comparison.” European Economic Review 41, 1201-1249.

31

Christiano, Lawrence, Martin Eichenbaum, and Charles Evans. (1999). “Monetary Policy

Shocks: What Have We Learned and to What End?” in Handbook of Macroeconomics 1A, edited by John Taylor.

Christiano, Lawrence, Martin Eichenbaum, and Charles Evans. (2001). “Nominal

Rigidities and the Dynamic Effects of a Shock to Monetary Policy.” Unpublished Manuscript.

Clarida, Richard, Jordi Gali, and Mark Gertler. (2000). “Monetary Policy Rules and

Macroeconomic Stability: Evidence and Some Theory.” Quarterly Journal of Economics 115, 147-80.

Doan, Thomas A. (2000). RATS Version 5 User’s Guide. Evanston, IL.: Estima. Eggertson, Gauti and Michael Woodford (2003). “The Zero Bound on Interest Rates and

Optimal Monetary Policy.” Unpublished manuscript. Estrella, Arturo, and Gikas A. Hardouvelis. (1991). “The Term Structure as a Predictor of

Real Economic Activity.” The Journal of Finance 46, 555-76. Evans, Charles, and David Marshall. (1998). “Monetary Policy and the Term Structure of

Nominal Interest Rates: Evidence and Theory.” Carnegie-Rochester Conference Series on Public Policy 49, 53-111.

Evans, Charles, and David Marshall. (2001). “Economic Determinants of the Nominal

Treasury Yield Curve.” Unpublished manuscript. Faust, Jon. (1999). “The Robustness of Identified VAR Conclusions About Money.”

Carnegie Rochester Conference Series 49, 207-244. Friedman, Milton. (1968). “The Role of Monetary Policy.” American Economic Review

58, 1-17. Gertler, Mark, and Simon Gilchrist. (1994). “Monetary Policy, Business Cycles, and the

Behavior of Small Manufacturing Firms.” Quarterly Journal of Economics 109, 309-40.

Hiraki, Takato, Noriyoshi Shiraishi, and Nobuya Takezawa. (1996). “Cointegration,

Common Factors, and the Term Structure of Yen Offshore Interest Rates.” Journal of

32

Fixed Income 6, 69-75. Ireland, Peter N. (1997). “A Small, Structural, Quarterly Model for Monetary Policy

Evaluation.” Carnegie-Rochester Conference Series on Public Policy 47, 83-108. Kahn, Charles M. and William Roberds (2002) “Payments Settlement under Limited

Enforcement: Private versus Public Systems.” Federal Reserve Bank of Atlanta Working Paper No. 2002-33.

Leeper, Eric, Christopher Sims, and Tao Zha. (1996). “What Does Monetary Policy

Do?” Brookings Papers on Economic Activity 2, 1-63. Litterman, Robert, and José Scheinkman. (1988). “Common Factors Affecting Bond

Returns.” The Journal of Fixed Income 1, 54-61. Rotemberg, Julio J. (1996). “Prices, Output, and Hours: An Empirical Analysis Based

on a Sticky Price Model.” Journal of Monetary Economics 37, 505-33. Miyao, Ryuzo. (2002). “The Effects of Monetary Policy in Japan.” Journal of Money,

Credit, and Banking 34, 376-392. Shioji, Etsuro. (1997). “Spanish Monetary Policy: A Structural VAR Analysis.”

Universitat Pompeu Fabra Working Paper 215. Sims, Christopher A. (1992). “Interpreting the Macroeconomic Time Series Facts: The

Effects of Monetary Policy.” European Economic Review 36, 975-1000. Singleton, Kenneth. (1994). “Yield Curve Risk in Japanese Government Bond Markets.”

Japanese Journal of Financial Economics 1, 5-31. Uhlig, Harald. (1999). “What are the Effects of Monetary Policy on Output? Results from

an Agnostic Identification Procedure.” CEPR Discussion Paper #2137. Zellner, A. (1971). An Introduction to Bayesian Inference in Econometrics. New York: Wiley. 1 We chose to estimate the VAR in levels. This produces consistent estimates even in situations where the

data are integrated or cointegrated. On the other hand, neither cointegration nor first difference

33

specifications produce consistent estimates under the alternative that the data is stationary in levels.

2 More details on the construction of the common factors is provided in section 3.

3 In Japan many ATM machines close on weekends and/or evenings.

4 In what follows, the month in which the shock arrives is labeled 0.

5 They describe the price level response as small, but not necessarily negative in early periods. This aspect

of the maintained hypothesis is explored in detail in (Braun and Shioji 2002) who find that relaxing the sign

restriction on prices produces a large, persistent and statistically significant price puzzle in Japanese data.

6 When the 12 month 36 month and 50 month yields are included instead of the common factors, the

condition number of the data matrix exceeds 30,000,000,000.

7 These confidence bands are likely to overstated the precision of these estimates due to our two step

estimation procedure.

8 The period t term premium response for a J period bond is: 1

, , ,10

1 J

t J t J t jj

tprem y yT

−

+=

= − ∑ .

9 The average impact response of monetary base is -0.35% with a standard error of .089% and the 1-month

rate response is -13 b.p. with a standard error of 2.1 b.p.

10 We also checked the robustness of this conclusion to the variants of the model reported in Section 4

below. However, if we condition on the estimated VAR coefficients and perform 50,000 replications, no

successful draws are found under the liquidity effect hypothesis.

11 See (Braun and Shioji 2002) for a detailed investigation of the plausibility of the two hypotheses.

12 For instance, the peak decline of the 5 year yield is –18.61 in the baseline case and –18.91 when R is

34

restricted to be non-negative.

13 Here the peak decline in the 5 year yield is –16.03.

14 The maximum decline in the 5 year yield is –18.1.

15 TIBOR and off-shore swap rate data are used because there is more liquidity at the various maturities in

these markets than in the Japanese Government Bond (JGB) market. Volume in the JGB market is thin at

many maturities and most transactions are concentrated in the market for the 10 year bell-weather bond.

yield1 month

rate factor 1 factor 22 months 0.999 0.032 -0.0743 months 0.996 0.052 -0.1186 months 0.986 0.109 -0.1889 months 0.971 0.158 -0.21612 months 0.963 0.191 -0.2324 months 0.907 0.359 -0.17236 months 0.851 0.439 -0.07848 months 0.793 0.494 0.01160 months 0.746 0.515 0.06685 months 0.660 0.542 0.148120 months 0.595 0.498 0.167

Table 1

Fraction of the conditional variance of each yield explained by the 1 month rate

and the two common factors

Variable 2 12 60 2 12 60Prices 0.30 0.99 5.38 ⎜ 0.28 6.52 16.07

(0.38) (1.15) (4.06) ⎜ (0.41) (5.43) (10.92)Industrial Production 2.89 3.19 3.92 ⎜ 2.77 11.37 15.52

(2.30) (2.05) (2.43) ⎜ (2.20) (8.74) (10.21)Monetary Base 57.93 46.79 17.37 ⎜ 41.61 20.99 25.16

(26.16) (12.78) (13.55) ⎜ (26.57) (12.76) (15.42)One month rate 28.56 10.05 6.71 ⎜ 45.17 41.90 28.17

(25.31) (12.72) (9.39) ⎜ (24.41) (16.00) (13.29)First common Factor 4.13 5.87 5.61 ⎜ 2.87 12.24 17.31

(2.33) (3.28) (3.49) ⎜ (1.98) (7.72) (9.39)Second Common Factor 11.29 11.00 7.44 ⎜ 6.49 15.46 23.26

(5.85) (4.87) (5.12) ⎜ (3.83) (5.80) (10.74)12-month yield 8.67 8.36 6.21 ⎜ 26.83 39.97 27.77

(11.80) (10.87) (8.45) ⎜ (12.52) (15.30) (12.92)5-year yield 4.75 7.61 6.85 ⎜ 13.97 35.07 27.56

(6.42) (8.68) (8.38) ⎜ (8.08) (14.16) (13.05)*Means across successful draws and standard deviations in parentheses

Step

Percentage of variance explained by monetary policy: liquidity effect

maintained hypothesis*

Percentage of variance explained by monetary policy: costly price

adjustment maintained hypothesis*

Table 2

Step

Variable 1 12 60 1 12 60Prices 2.69 4.98 13.12 ⎢ 2.62 6.86 11.71

(2.40) (4.72) (9.54) ⎢ (2.50) (5.42) (8.91)Industrial Production 1.54 6.37 15.00 ⎢ 2.12 8.51 13.83

(2.10) (6.52) (10.68) ⎢ (2.38) (10.47) (8.27)Monetary Base 12.69 10.50 8.17 ⎢ 8.26 35.07 27.56

(11.80) (11.31) (7.42) ⎢ (7.46) (14.16) (13.05)Monetary Policy 26.83 39.97 27.77 ⎢ 13.97 35.07 27.56

(12.52) (15.30) (12.92) ⎢ (8.08) (14.16) (13.05)First Common Factor 41.80 16.27 11.55 ⎢ 72.31 29.07 18.49

(7.49) (8.62) (7.24) ⎢ (7.52) (10.25) (8.73)Second Common Factor 14.45 21.91 24.39 ⎢ 0.71 9.63 19.20

(3.38) (11.20) (13.20) ⎢ (0.38) (7.20) (12.04)*Means across successful draws and standard deviations in parentheses

Table 3

Step

Percentage of variance in 5-year yield explained by VAR shocks under costly

price adjustment hypothesis*

Percentage of variance in 12-month yield explained by VAR shocks under

costly price adjustment hypothesis*Step

Variable 1 12 60 1 12 60Prices 0.41 2.78 6.48 1.20 3.51 10.06

(0.58) (2.74) (5.48) (1.15) (2.60) (8.75)Industrial Production 0.77 16.06 13.82 3.55 7.10 12.29

(1.01) (8.50) (8.26) (2.45) (5.71) (8.12)Monetary Base 0.79 10.04 8.72 0.57 5.56 22.46

(0.95) (6.99) (8.17) (0.82) (4.70) (14.44)One month rate 8.04 10.89 9.28 7.25 23.65 24.78

(3.96) (8.14) (7.23) (3.12) (11.37) (13.73)*Means across successful draws and standard deviations in parentheses

Step

Percentage of variance in each variable explained by first

common factor under costly price adjustment hypothesis*

Percentage of variance in each variable explained by second common factor under sticky

price hypothesis*

Table 4

Step

Figu

re 1

base

line

spec

ifica

tion

Res

pons

e of

VA

R v

aria

bles

tosh

ock

to m

onet

ary

polic

yLi

quid

ity e

ffec

t mai

ntai

ned

hypo

thes

is50

0 ou

terlo

op d

raw

s, 10

0 in

nerlo

op d

raw

sba

selin

e sp

ecifi

catio

n

Res

pons

e of

VA

R v

aria

bles

tosh

ock

to m

onet

ary

polic

yC

ostly

pric

e ad

just

men

t mai

ntai

ned

hypo

thes

is50

0 ou

terlo

op d

raw

s, 10

0 in

nerlo

op d

raw

s

Res

pons

e of

CPI

-0.0

015

-0.0

01-0

.000

500.

0005

0.00

1

14

710

1316

1922

Res

pons

e of

Y

-0.0

06-0

.004

-0.0

0200.

002

0.00

4

14

710

1316

1922

Res

pons

e of

M

-0.0

1-0

.008

-0.0

06-0

.004

-0.0

0200.

002

14

710

1316

1922

Res

pons

e of

R

-20

-100102030

14

710

1316

1922

Res

pons

e of

F1

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

F2

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

lev

el

-20

-1001020

14

710

1316

1922

Res

pons

e of

slo

pe

-15

-10-50510

14

710

1316

1922

Res

pons

e of

cur

vatu

re

-10-50510

14

710

1316

1922

Res

pons

e of

real

rate

-90

-401060

14

710

1316

1922

Res

pons

e of

CPI

-0.0

03-0

.002

-0.0

0100.

001

14

710

1316

1922

Res

pons

e of

Y

-0.0

1

-0.0

050

0.00

5

14

710

1316

1922

Res

pons

e of

M

-0.0

1-0

.0050

0.00

50.

01

14

710

1316

1922

Res

pons

e of

R

-40

-20020

14

710

1316

1922

Res

pons

e of

F1

-0.0

3-0

.02

-0.0

100.

010.

02

14

710

1316

1922

Res

pons

e of

F2

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

lev

el

-40

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

slo

pe

-10-5051015

14

710

1316

1922

Res

pons

e of

cur

vatu

re

-10-5051015

14

710

1316

1922

Res

pons

e of

real

rate

-90

-401060

14

710

1316

1922

Figu

re 2

base

line

spec

ifica

tion

Res

pons

e of

VA

R y

ield

cur

ve to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

base

line

spec

ifica

tion

Res

pons

e of

VA

R y

ield

cur

ve to

shoc

k to

mon

etar

y po

licy

Liqu

idity

eff

ect m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

Res

pons

e of

6-m

onth

rate

-20

-1001020

14

710

1316

1922

Res

pons

e of

12-

mon

th ra

te

-30

-20

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

12-

mon

th te

rm p

rem

ium

-20

-1001020

14

710

1316

1922

Res

pons

e of

5-y

ear r

ate

-20

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

5-y

ear t

erm

pre

miu

m

-20

-1001020

14

710

1316

1922

Res

pons

e of

6-m

onth

term

pre

miu

m

-10-50510

14

710

1316

1922

Res

pons

e of

6-m

onth

rate

-40

-20020

14

710

1316

1922

Res

pons

e of

12-

mon

th ra

te

-40

-30

-20

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

12-

mon

th te

rm p

rem

ium

-20

-1001020

14

710

1316

1922

Res

pons

e of

5-y

ear r

ate

-40

-30

-20

-10010

13

57

911

1315

1719

2123

Res

pons

e of

5-y

ear t

erm

pre

miu

m

-40

-20020

14

710

1316

1922

Res

pons

e of

6-m

onth

term

pre

miu

m

-10-50510

14

710

1316

1922

Figu

re 3

base

line

varia

bles

and

yen

-dol

lar e

xcha

nge

rate

(ord

ered

5th

)

Res

pons

e of

VA

R v

aria

bles

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

base

line

varia

bles

and

pco

m (o

rder

ed 3

rd)

Res

pons

e of

VA

R v

aria

bles

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

base

line

varia

bles

and

PO

IL (o

rder

ed 3

rd)

Res

pons

e of

VA

R v

aria

bles

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

Res

pons

e of

CPI

-0.0

02

-0.0

010

0.00

1

14

710

1316

1922

Res

pons

e of

Y

-0.0

1

-0.0

050

0.00

5

14

710

1316

1922

Res

pons

e of

PC

OM

-0.0

03-0

.002

-0.0

0100.

001

0.00

2

14

710

1316

1922

Res

pons

e of

M

-0.0

1

-0.0

050

0.00

5

14

710

1316

1922

Res

pons

e of

R

-30

-20

-10010

13

57

911

1315

1719

2123

Res

pons

e of

F1

-0.0

3-0

.02

-0.0

100.

010.

02

14

710

1316

1922

Res

pons

e of

F2

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

slo

pe

-10-5051015

13

57

911

1315

1719

2123

Res

pons

e of

cur

vatu

re

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

rea

l rat

e

-50

-30

-10103050

13

57

911

1315

1719

2123

Res

pons

e of

CPI

-0.0

02

-0.0

010

0.00

1

14

710

1316

1922

Res

pons

e of

Y

-0.0

1-0

.0050

0.00

50.

01

14

710

1316

1922

Res

pons

e of

PO

IL

-0.0

6-0

.04

-0.0

200.

020.

04

13

57

911

1315

1719

2123

Res

pons

e of

M

-0.0

1-0

.0050

0.00

50.

01

14

710

1316

1922

Res

pons

e of

R

-40

-30

-20

-10010

13

57

911

1315

1719

2123

Res

pons

e of

F1

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

F2

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

slo

pe

-10-5051015

13

57

911

1315

1719

2123

Res

pons

e of

cur

vatu

re

-10-5051015

13

57

911

1315

1719

2123

Res

pons

e of

rea

l rat

e

-60

-40

-200204060

13

57

911

1315

1719

2123

Res

pons

e of

CPI

-0.0

03-0

.002

-0.0

0100.

001

14

710

1316

1922

Res

pons

e of

Y

-0.0

1

-0.0

050

0.00

5

14

710

1316

1922

Res

pons

e of

M

-0.0

1-0

.0050

0.00

50.

01

14

710

1316

1922

Res

pons

e of

R

-30

-20

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

YEN

DO

L

-0.0

3-0

.02

-0.0

100.

010.

02

13

57

911

1315

1719

2123

Res

pons

e of

F1

-0.0

3-0

.02

-0.0

100.

010.

02

14

710

1316

1922

Res

pons

e of

F2

-0.0

2-0

.010

0.01

0.02

14

710

1316

1922

Res

pons

e of

slo

pe

-10-5051015

13

57

911

1315

1719

2123

Res

pons

e of

cur

vatu

re

-1001020

13

57

911

1315

1719

2123

Res

pons

e of

rea

l int

eres

t rat

e

-60

-40

-200204060

13

57

911

1315

1719

2123

Figu

re 4

base

line

varia

bles

and

PO

IL (o

rder

ed 3

rd)

Res

pons

e of

yie

ld c

urve

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

base

line

varia

bles

and

yen

-dol

lar e

xcha

nge

rate

(ord

ered

5th

)

Res

pons

e of

yie

ld c

urve

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

base

line

varia

bles

and

pco

m (o

rder

ed 3

rd)

Res

pons

e of

yie

ld c

urve

to

shoc

k to

mon

etar

y po

licy

Cos

tly p

rice

adju

stm

ent m

aint

aine

d hy

poth

esis

500

oute

rloop

dra

ws,

100

inne

rloop

dra

ws

Res

pons

e of

12-

mon

th ra

te

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

12-

mon

th te

rm

prem

ium

-20

-1001020

14

710

1316

1922

Res

pons

e of

5-y

ear r

ate

-30

-20

-10010

14

710

1316

1922

Res

pons

e of

5-y

ear t

erm

pre

miu

m

-40

-2002040

14

710

1316

1922

Res

pons

e of

6-m

onth

rate

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

6-m

onth

term

pre

miu

m

-10-50510

13

57

911

1315

1719

2123

Res

pons

e of

12-

mon

th ra

te

-30

-20

-10010

14

710

1316

1922

Res

pons

e of

12-

mon

th te

rm p

rem

ium

-20

-1001020

14

710

1316

1922

Res

pons

e of

5-y

ear r

ate

-30

-20

-10010

14

710

1316

1922

Res

pons

e of

5-y

ear t

erm

pre

miu

m

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

6-m

onth

rate

-30

-20

-10010

14

710

1316

1922

Res

pons

e of

6-m

onth

term

pre

miu

m

-10-50510

14

710

1316

1922

Res

pons

e of

6-m

opnt

h te

rm p

rem

ium

-10-50510

14

710

1316

1922

Res

pons

e of

12-

mon

th ra

te

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

12-

mon

th te

rm p

rem

ium

-15

-10-5051015

13

57

911

1315

1719

2123

Res

pons

e of

5-y

ear r

ate

-30

-20

-10010

14

710

1316

1922

Res

pons

e of

5-y

ear t

erm

pre

miu

m

-30

-20

-1001020

14

710

1316

1922

Res

pons

e of

6-m

onth

rate

-30

-20

-1001020

14

710

1316

1922